SDM

We chose six species distribution models: BIOCLIM, DOMAIN, MAHAL (Mahalanobis distance), RF (random forests), MAXENT (maximum entropy), and SVM (support vector machine). These six SDMs are widely used in academic research and species conservation [16] , [17] , [33] , [34] .
The BIOCLIM model uses environmental data of all known species distribution points and can determine the range of weather conditions suitable for species occurrence. The percentile distribution of every climatic variable within each grid in the species distributions zone is used for multivariate analysis. If the ranges of all climatic variables in the grid are within boundaries appropriate for that species, the BIOCLIM model indicates that this place is suitable [8] , [17] .
The DOMAIN uses a point-to-point similarity metric based on the Gower distance, which is a method for creating a distance matrix from a set of characteristics of species. DOMAIN can assign a classification value of habitat suitability index to each potential site based on its proximity in environmental space to the most similar positive occurrence location [38] . Then, a threshold value of suitability is chosen to determine the distribution boundaries of species’ ecological niche.
The MAHAL model is based on Mahalanobis distance (MD). MD considers the variables correlations in the data set without depending on the scale of measurements. The method ranks the potential sites through their MD to a vector, which can express the mean environmental values of all recorded environmental factors. A certain distance threshold can act as the ecological niche boundaries. These algorithm generate an elliptic envelope which can explicitly explain the possible interrelations between these environmental factors [21] (link).
The RF, a classification and regression tree model, is a combination of tree predictors where every tree can depend on the values of a random vector sampled independently with the same distribution for all trees in the forest [39] , [40] .
MAXENT is based on a machine learning algorithm called maximum entropy, and is based on the principle that species without ecological constraints will spread as far as possible with a distribution as close as possible to uniform [41] .
The SVM is a machine-learning method that belongs to a family of generalized linear classifiers. The principle of SVM is the Vapnik Chervonenkis (VC) dimension and structural risk minimization theory [42] . The SVM model can find the most reasonable way between species adaptability and complexity to yield the most likely distribution according to the limited sample information [43] .
Each of the SDMs was operated with strictly following the modeling technique and using the same 13 environmental variables. Modeling data, advantages and disadvantages were listed in Table S2. We chose “R” as the computing platform and the dismo package to simulate species distribution [44] , [45] .

An ensemble of projections of SDM was obtained for the 11,495 selected species. The ensemble included projections with Generalized Additive Models, Boosting Regression Trees, Generalized Linear Models and Random Forests. Models were calibrated for the baseline period using 70% of observations randomly sampled from the initial data and evaluated against the remaining 30% data using the true skill statistic (TSS^{29 (link)}). Presence data were randomly drawn from the gridded range maps. For absences, we considered data in a reasonable buffer around the presence data to avoid having over-optimistic predictive accuracies^{30 (link)}. To be consistent with assumed realistic dispersal distances (see next paragraph) and in line with previous analyses, we selected absence data in 2000, 3000, and 4000 km buffer around amphibian, mammal and bird species ranges, respectively. This analysis was repeated four times, thus providing a four-fold internal cross-validation of the models (biomod package^{15 (link)}). The quality of the models was ‘very high’ to ‘excellent’ with an average TSS of 0.83 (Supplementary Fig. 2). Since the quality of the models strongly affect projection uncertainties, we tested three thresholds below which models were removed from projection ensembles. We used a threshold of TSS = 0.4, which is usually considered a minimum for retaining reliable models^{29 (link)}. We also used thresholds of 0.6 and 0.7, and investigated their effects on uncertainties of species-based sensitivity metrics. For all further analyses, the most drastic threshold (TSS = 0.7) was then used (red lines in Supplementary Fig. 2).
For each species, all calibrated models (4 SDMs × 4 repetitions) were then used to project the potential distribution of each species under both current and projected future climatic conditions (Supplementary Fig. 1).

To study the effect of altering the geographic extent of the region modeled (i.e., the area in which the model is trained) on predicted species’ distributions, we first generated SDMs using a wide geographical extent including the entire region of Fennoscandia and north-western Russia (map shown in Figure 3B). Due to the paucity of Falco records in north-western Russia, using a full geographic extent of the region modeled may characterize the realized distribution of species in the genus poorly. Including large areas increases the chance that the model samples pseudo-absences in areas that have suitable conditions for the species but are falsely classified as unsuitable because the species has not been properly sampled in that region [8] . Indeed, choosing the correct extent is not a trivial task since the values where occurrence data are lacking are taken as pseudo-absences that are meant to provide a comparative data set to establish the conditions where a species may occur. If large extents with great environmental variation are selected, predictive models will be dominated by parameters that serve to coarsely discriminate regional conditions and weaken the ability to tease out fine-scale conditions determining presence or absence of species [27] . On the other hand, using a restricted region for selection of pseudo-absences can be a serious error when fitting models to project potential effects of climate change [37] , since future environmental conditions may not be represented. Since occurrence data for our model species was lacking for north-western Russia, we used Fennoscandia, which accurately mirrored the distribution of the occurrence data of the species (map shown in Figure 3A).

Results for hierarchical cluster analysis (HCA) of samples and metabolites are presented as heatmaps with dendrograms. Pearson correlation coefficients (PCC) between samples were calculated using the cor function in R (accessed on: 10 December 2021) and presented as only heatmaps. Both HCA and PCC were carried out using the R package pheatmap. For HCA, normalized signal intensities of metabolites (unit variance scaling) are visualized as a color spectrum. Principal component analysis (PCA) was performed with the statistics function prcomp within R. The data were unit variance scaled before PCA. Significantly regulated metabolites (SRMs) between groups were determined by VIP ≥ 1 and absolute Log₂FC (fold change) ≥ 1. VIP values were extracted from OPLS-DA results, which also contained score plots and permutation plots, and was generated using R package MetaboAnalystR. The data were log transformed (log₂) and mean centered before OPLS-DA. To avoid overfitting, a permutation test (200 permutations) was performed. Identified metabolites were annotated using the KEGG compound database (accessed on: 10 December 2021) and annotated metabolites were then mapped to the KEGG pathway database (accessed on: 10 December 2021). Pathways with the SDMs mapped were subjected to MSEA (metabolite sets enrichment analysis), and their significance was determined using p-values from a hypergeometric test.

We used species distribution models which envisage potentially suitable habitat for species based on variables supportive to species occurrences [33 (link)]. Numerous model types are available for developing SDMs; however, maximum entropy (e.g., Maxent) is one of most effective tools for species having limited spatial presence-only data [34 ,35 (link),36 (link)]. Maximum entropy forecasts the species probability distribution of random events with uniform and high stability [37 (link),38 (link)].
We determined potential habitat layers by first selecting 19 bioclimatic variables at 30 arc-second (about 1 km²) resolution from Worldclim [39 (link)] and extracted aspect from a digital elevation model (about 1-km² resolution; Shuttle Radar Topography Mission (SRTM); https://ita.cr.usgs.gov, accessed on 28 November 2022) (Supplementary Table S1). We selected eight environmental variables including mean annual temperature (Bio_1), mean diurnal range (Bio_02), iso-thermality (Bio_3), annual precipitation (Bio_12), precipitation of driest month (Bio_14), precipitation seasonality (Bio_15), precipitation of coldest quarter (Bio_19), and aspect (see Figure 2) based on jackknife analysis, which uses a leave-one-out approach to estimate variable importance (Supplementary Materials Table S1; Figure S1). We excluded highly correlated (|r| > 0.70) variables to reduce the potential for model overfitting [40 (link)]. We used maximum entropy modeling software (Maxent version 3.3.3 K) [36 (link)] to estimate the potential nesting distribution of Egyptian vultures.
We ran Maxent models using 75% presence data for calibration and the remaining 25% for model validation [41 (link)] using the bootstrapping procedure with 10 replications and a maximum of 500 iterations. We used the default settings for the number of background samples and auto features for model construction. We evaluated model validation and accuracy using the area under the curve (AUC) of the receiver operating characteristics [42 ]. However, because of criticism in the use of AUC for model evaluation [43 (link)], we also evaluated model performance using the true skill statistic (TSS), sensitivity, and specificity [44 (link)]. The TSS values range from −1 to 1, with values approaching 1 indicating good model performance and values <0 indicating performance no better than random. We converted habitat into binary form (suitable or unsuitable) using the 10th-percentile training presence logistic threshold [45 (link)] and used the corresponding raster layer to identify suitable potential nesting habitat for Egyptian vultures.
We used predicted suitable nesting habitat to identify whether the model correctly predicted nesting habitat as well as existing factors influencing the occurrence of Egyptian vulture nests along the Madi River corridor from Damauli, Tanahu to Tanting, Kaski. This area supports abundant Egyptian vultures and we assumed this abundance reflected their residence in this area, where cliffs and forests occur near human settlements. We established 35 stations at 3-km intervals along 135 km of the Madi River. Each station was 500 m × 500 m and within each we established five 100 m × 100 m plots, four at the corners and one at the center. During October 2020–February 2021, we searched each plot for 3 h, recording the occurrence of Egyptian vulture nests and the presence of Egyptian vultures roosting or feeding. We recorded the occurrence of feeding vultures while sitting quietly at the corner of each plot, and for observations of nesting and roosting we searched throughout the plots. From the center of each plot we recorded the elevation using GPS, and measured the distance to the nearest forest, area of agricultural land, water, and human settlement. We measured the nearest distance to human settlement using a GIS if the area had more than one house and were used for living.